The lower extent of error bars is not visible in (B), (D), (G), or (E). Avoid “dynamite plots.”
Bar charts should be avoided; a 1D scatterplot of each dataset in (D), (G), and (E) would be clearer. In (B) a line plot with concentration on the \(x\)-axis would allow direct comparison of A16 and A18 and show trends.
We can’t see whether there is a difference between the effects of the two concentrations of A16 in (D), or between A16 and A18 in (B); use A table of contrasts.
Place things to be compared by the reader next to each other where possible (E).
UMAP plots (B, E) are highly manipulable and clustering/placement does not necessarily reflect objective measures.
Unpleasant colour choices in (C); there is room for aesthetic improvement.
The proportion plot in (C) does not give information on absolute number, only proportion; a proportional areas plot spanning all clusters would more honestly represent the data.
Heatmap text is too small to read comfortably; is there too much data here?